I used to catch the bus every day at I-405 & Coal Creek Parkway in Bellevue, WA. http://g.co/maps/ack5b

Not a perfect answer, but a survey-of-the-literature paper from 2004 suggests the short-term elasticity from transit service as 0.5 to 0.7 and long-term elasticity as 0.7 to 1.1. Of course, "service" can be more than just frequency. The paper is here: http://www.nctr.usf.edu/jpt/pdf/JPT%207-2%20Litman.pdf

Since nobody's yet answered Jarrett's extra credit question, I'll take a stab. The first part, where should the stops be located on the second parallel line, is easy. They should be spaced exactly between the stops on the first line, to maximize the coverage area for the second line filling in the coverage gaps from the first line. As to how far apart the lines can be before this stops mattering...if we take the 400m maximum walking distance with the stops ever 400m (per the Australian/European standard), then a parallel line spaced 600m from the first line would perfectly fill in the coverage area gaps. As the line gets further away, the overlap diminishes until by 800m the lines aren't working together at all. The biggest learning to me from the last part is that any closer than 600m and the lines are wastefully covering the same area. That would mean crosstown buses in Manhattan are optimally spaced about 7 or 8 blocks apart.

Thanks Jarrett for putting numbers behind the difficulties of Dial-A-Ride. I had the same thought as anonymouse...a Dial-A-Ride service supported by smartphone technology. To my knowledge there is no good implementation of this. Perhaps the best execution would be to ask people requesting a ride to walk the few blocks to their nearest arterial. That way the service would be something closer to a dynamically-optimized fixed route system.

As a former Seattle resident, your description of the high ridership plan sounds pretty good to me. Could the empty-running coverage-oriented buses be replaced with some sort of dial-a-ride system running full?